ind = split.split(labels, groups=labels['SourceReportName'])
trn_ind, val_ind = next(ind)
trn_df = labels.loc[trn_ind, ]
val_df = labels.loc[val_ind, ]
train_gen = ImageFrameGenerator(
rotation_range=180,
width_shift_range=0.2,
height_shift_range=0.2,
shear_range=0.1,
zoom_range=0.2,
horizontal_flip=True,
vertical_flip=True,
fill_mode='nearest')
test_gen = ImageFrameGenerator()
trn_itr = train_gen.flow_from_frame(data_path, trn_df, 'basefile', ['split0_123', 'split01_23', 'split012_3'],
target_size=(256, 320), label_types = ['categorical']*3, batch_size=batch_size)
val_itr = test_gen.flow_from_frame(data_path, val_df, 'basefile', ['split0_123', 'split01_23', 'split012_3'],
target_size=(256, 320), label_types = ['categorical']*3, batch_size=batch_size, shuffle=False)
from keras.applications.inception_v3 import InceptionV3
base_model = InceptionV3(weights='imagenet', include_top=False)
x = base_model.output
x = GlobalAveragePooling2D()(x)
output1 = Dense(2, activation='softmax', name='split0_123')(x)
output2 = Dense(2, activation='softmax', name='split01_23')(x)
output3 = Dense(2, activation='softmax', name='split012_3')(x)
model = Model(inputs=base_model.input, outputs=[output1, output2, output3])
base_model = InceptionV3(weights='imagenet', include_top=False)
x = base_model.output
x = GlobalAveragePooling2D()(x)
output1 = Dense(2, activation='softmax', name='split0_123')(x)
output2 = Dense(2, activation='softmax', name='split01_23')(x)
output3 = Dense(2, activation='softmax', name='split012_3')(x)
model = Model(inputs=base_model.input, outputs=[output1, output2, output3])
tst_gen = ImageFrameGenerator()
tst_itr = tst_gen.flow_from_frame(data_path,
video_labels,
'filename',
['split0_123', 'split01_23', 'split012_3'],
target_size=(256, 320),
label_types=['categorical'] * 3,
batch_size=batch_size,
shuffle=False)
thresh = 0.5
auc01_23 = []
sensi01_23 = []
speci01_23 = []
ppv01_23 = []
npv01_23 = []
acc0 = []
acc1 = []
acc2 = []
acc3 = []
kappa_linear = []
trn_ind, val_ind = next(ind)
trn_df = labels.loc[trn_ind, ]
val_df = labels.loc[val_ind, ]
train_gen = ImageFrameGenerator(rotation_range=180,
width_shift_range=0.2,
height_shift_range=0.2,
shear_range=0.1,
zoom_range=0.2,
horizontal_flip=True,
vertical_flip=True,
fill_mode='nearest')
test_gen = ImageFrameGenerator()
train_itr = train_gen.flow_from_frame(data_path,
trn_df,
'basefile', ['Mayo_score'],
target_size=(256, 320),
batch_size=batch_size)
val_itr = test_gen.flow_from_frame(data_path,
val_df,
'basefile', ['Mayo_score'],
target_size=(256, 320),
batch_size=batch_size)
from keras.applications.inception_v3 import InceptionV3
base_model = InceptionV3(weights='imagenet', include_top=False)
x = base_model.output
x = GlobalAveragePooling2D()(x)
output1 = Dense(1, activation='linear', name='Mayo_score')(x)
val_size = int(len(df) * 0.125 / tot_bs) * tot_bs
trn_size = int((len(df) - val_size) / tot_bs) * tot_bs
trn_df, val_df = train_test_split(
df,
test_size=val_size,
train_size=trn_size,
stratify=df[['CD_Active_AnyLocation', 'Fistula_Any', 'Abscess_any']],
random_state=24)
gen = ImageFrameGenerator()
trn_itr = gen.flow_from_frame(
path + 'low_resolution/',
trn_df,
'filename', ['CD_Active_AnyLocation', 'Fistula_Any', 'Abscess_any'],
target_size=(224, 224, 128),
color_mode='3d',
batch_size=batch_size * G,
shuffle=True)
val_itr = gen.flow_from_frame(
path + 'low_resolution/',
val_df,
'filename', ['CD_Active_AnyLocation', 'Fistula_Any', 'Abscess_any'],
target_size=(224, 224, 128),
color_mode='3d',
batch_size=batch_size * G,
shuffle=False)
from dense3dnet import Dense3DNet
blocks = [6, 12, 24, 16]
trn_df = labels.loc[trn_ind, ]
val_df = labels.loc[val_ind, ]
train_gen = ImageFrameGenerator(rotation_range=180,
width_shift_range=0.2,
height_shift_range=0.2,
shear_range=0.1,
zoom_range=0.2,
horizontal_flip=True,
vertical_flip=True,
fill_mode='nearest')
test_gen = ImageFrameGenerator()
trn_itr = train_gen.flow_from_frame(
data_path,
trn_df,
'basefile',
['Inflamm_Mayo_0', 'Inflamm_Mayo_1', 'Inflamm_Mayo_2', 'Inflamm_Mayo_3'],
target_size=(256, 320),
batch_size=batch_size,
shuffle=True)
val_itr = test_gen.flow_from_frame(
data_path,
val_df,
'basefile',
['Inflamm_Mayo_0', 'Inflamm_Mayo_1', 'Inflamm_Mayo_2', 'Inflamm_Mayo_3'],
target_size=(256, 320),
batch_size=batch_size,
shuffle=False)
from keras.applications.inception_v3 import InceptionV3
base_model = InceptionV3(weights='imagenet', include_top=False)
trn_ind, val_ind = next(ind)
trn_df = labels.loc[trn_ind, ]
val_df = labels.loc[val_ind, ]
train_gen = ImageFrameGenerator(rotation_range=180,
width_shift_range=0.2,
height_shift_range=0.2,
shear_range=0.1,
zoom_range=0.2,
horizontal_flip=True,
vertical_flip=True,
fill_mode='nearest')
test_gen = ImageFrameGenerator()
trn_itr = train_gen.flow_from_frame(data_path,
trn_df,
'basefile', ['Mayo_score'],
label_types=['categorical'],
target_size=(256, 320),
batch_size=batch_size)
val_itr = test_gen.flow_from_frame(data_path,
val_df,
'basefile', ['Mayo_score'],
label_types=['categorical'],
target_size=(256, 320),
batch_size=batch_size,
shuffle=False)
from keras.applications.inception_v3 import InceptionV3
base_model = InceptionV3(weights='imagenet', include_top=False)
x = base_model.output
x = GlobalAveragePooling2D()(x)